Strange values in convolution (using FFT through gpu)

I want to realize a code in which i use the convolution for GPU computing. So I'm studying all the characteristics of convulution and its application for GPU. I've also read this question https://it.mathworks.com/matlabcentral/answers/38066-difference-between-conv-ifft-fft-when-doing-convolution (very useful):

So look at this code:

len=100;
g=randi(50,1,len);
h=randi(50,1,len);
z=conv(g,h);

s=len+len-1;        % it is related to the padding
l=gpuArray([g zeros(1,s-len)]);                       
m=gpuArray([h zeros(1,s-len)]);                       
z_2=ifft(fft(l).*fft(m));                    
z_2=gather(z_2);

Here there are the first components of the results :

z = 8  185  931  1623  2278  3859  4270  5559  7330  6475  7549  7862  7893  10386  10595  11300

z_2= 7,99999999998245 - 3,51000970332467e-12i  184,999999999994 - 1,13000769162752e-12i  930,999999999987 + 2,84583190541624e-12i  1622,99999999998 + 4,19583394168701e-14i  2277,99999999999 - 7,29777384031118e-12i  3858,99999999999 - 6,95099718695151e-12i  4269,99999999998 - 7,31342157267749e-12i  5558,99999999998 + 3,52422064534209e-13i  7329,99999999999 + 5,34843173262111e-12i  6474,99999999998 - 6,95563862181934e-12i  7548,99999999998 - 4,98451114397262e-12i  7862,00000000000 - 1,06265252202149e-11i  7892,99999999998 - 1,42314382643217e-11i  10386,0000000000 + 2,28547111462919e-12i  10595,0000000000 + 8,34669775247101e-12i  11300,0000000000 + 4,30036588495909e-12i

 Why is there the imag part in the second "version" ?